Code Generators for MySQL Plugins and User Defined Functions. Hartmut Holzgraefe

Transcription

1 Code Generators for MySQL Plugins and User Defined Functions Hartmut Holzgraefe

2 Why UDFs / Plugins? to add functionality not available elsewhere to build upon functionality already available in a C/C++ library to get better performance by using native code

3 Why a code generator? less steep learning curve faster development, focus on actual functionality API changes can be dealt with in one central place automate boring tasks

4 Aspects of building a UDF or Plugin Requirements & Dependencies Build Infrastructure API Infrastructure Implementation of actual functionality Testing Documentation Packaging & Distribution

5 Generator Infrastructure Generator Requirements PHP 5.x PEAR ( PHP Extension & Application Repository ) PEAR Packages: CodeGen & CodeGen_MySQL CodeGen_MySQL_UDF CodeGen_MySQL_Plugin CodeGen_Drizzle

6 Build Infrastructure Requirements for building: MySQL Development Packages sometimes full MySQL Server Soure C/C++ compiler and make autoconf, automake, libtool or on Windows: CMake (w.i.p.)

7 Getting started on Debian/Ubuntu apt-get build-dep mysql-server apt-get install php5-cli pear pear install CodeGen_MySQL_UDF pear install CodeGen_MySQL_Plugin pear install CodeGen_Drizzle

8 Build Infrastructure #2 Things taken care of: configure script Makefile make targets for: make all make test make clean make dist / distcheck make html / pdf

9 Getting Started A minimal UDF specification file: <?xml version="1.0"?> <udf name="dummy"> </udf> and code is generated from it using: udf-gen dummy.xml

10 Generated Files twice -- AUTHORS -- CMakeLists.txt -- ChangeLog -- INSTALL -- Makefile.am -- NEWS -- README -- configure.ac -- twice.c -- m4/ -- ax_compare_version.m4 -- mysql.m4 -- manual.xml -- mysql.m4 -- tests/ -- create_functions.inc -- drop_functions.inc -- r/ -- t/ `-- test.sh `-- udf_twice.h

11 A 'real' Example <?xml version="1.0"?> <udf name="twice"> <function name="twice" returns="int"> <param name="val" type="int"/> <code> return val * 2; </code> </function> </udf>

12 Compiling the generated Code $ udf-gen twice.xml $ cd example $./configure --with-mysql=... --libdir=... $ make

13 Instaling the generated Function Install the UDF libary sudo make install Register the function CREATE FUNCTION twice RETURNS INTEGER SONAME "twice.so"; Test the function SELECT twice(3); -- result: 6

14 Aggregate functions <function name="my_sum" type="aggregate" returns="int"> <param name="val" type="int"/> <data> <element name="sum" type="long" default="0"/> </data> <start> data->sum = 0; </start> <add> data->sum += val; </add> <result> return data->sum; </result> </function>

15 Automated Tests Automated tests that use the mysqltest tool can be generated using a simplified syntax: <test name= twice > <code> SELECT twice(3); </code> <result> twice(3); 6 </result> </test>

16 Invoking the Tests Both the mysqltest and mysqld binaries have to be in $PATH invoke tests with make test $ PATH=$PATH/usr/local/mysql/bin $ PATH=$PATH/usr/local/mysql/libexec $ export PATH $ make test *** testing *** [twice] ok

17 Packing a Source Release To pack the generated source for distribution use either make dist or make distcheck make dist creates a source tarball make distcheck creates a source tarball, unpacks it in a temporary directory and checks that building it works (e.g. no files left out when packing) Documentation can be generated using make html and make pdf if you have docbook-utils installed

18 Things not shown here Author, Copyright and License Information Library and Header File Dependencies Adding C/C++ code files Lex/Flex and Yacc/Bison support Windows Builds using CMake DBUG functions

19 Plugins allow to extend the server using shared libraries and well defined registration APIs currently supported plugin types are deamons storage engines fulltext parsers INFORMATION_SCHEMA tables Audit Replication Authentication

20 Building a Plugin Generating a plugin works like generating a UDF: mysql-plugin-gen my-plugin.xml To configure the plugin use configure --with-mysql=... --libdir=... or if the plugin requires server internals configure --with-mysql-src=... --libdir=...

21 Installing a Plugin first install the plugin library sudo make install then for each plugin defined in the library do INSTALL PLUGIN plugin_name SONAME plugin_library.so ; to uninstall use UNINSTALL PLUGIN;

22 A Daemon Plugin <daemon name="dummy"> <summary>minimal example plugin</summary> <init>fprintf(stderr, daemon started\n );</init> <deinit>fprintf(stderr, deamon stopped\n );</deinit> </daemon> </plugin> <plugin name="daemon_example"> <maintainer> <name>hartmut Holzgraefe</name> </maintainer> <license>gpl</license> <release> <version>0.0.1</version> <date> </date> <state>alpha</state> <notes>just an experiment</notes> </release>

23 Adding server variables <statusvar type="int" name="the_answer" init="42" /> SHOW STATUS LIKE 'THE_ANSWER'; Variable_name Value THE_ANSWER 42

24 An INFORMATION_SCHEMA Plugin <deps language="c++"> <src/> <header= sql/mysql_priv.h /> </deps> <infoschema name="is_dummy"> <summary>minimal information_schema plugin</summary> <statusvar type="int" name="dummy_count" init="0"/> <field name="foo_int" type="long"/> <field name="foo_string" type="string"/> <code> <[![CDATA TABLE* table= tables->table; table->field[field_foo_int]->store(23); table->field[field_foo_string]->store("foobar", 7, system_charset_info); schema_table_store_record(thd, table); table->field[field_foo_int]->store(42); table->field[field_foo_string]->store("barfoo", 7, system_charset_info); schema_table_store_record(thd, table); dummy_count++; return 0; ]]> </code> </infoschema>

25 A Fulltext Parser Plugin <fulltext name="dummy"> <summary>minimal example plugin</summary> <parser> <code> <![CDATA[ char *doc; int result; doc = strndup(param->doc, param->length); result = param->mysql_parse(param, param->doc, param->length); free(doc); return result; ]]> </code> </parser> </fulltext>

26 Things not covered (here)... Storage engine plugins Fulltext plugin details Status variable datatypes and callbacks Command line options header file and library dependencies...

27 Conclusion The generator tools save time: less code to write less things to learn no debugging of API infrastructure code on API changes just re-generate with current generator

28 Conclusion #2 Things already implemented using these tools: UDF_REGEXP UDF_ORA I_S.MASTER_STATUS and I_S.SLAVE_STATUS a Fulltext Parser for gzip compressed data

29 Online Resources Trac/Wiki PEAR Packages UDF projects